Kove:SDM was the first to deliver virtualized memory

After working to solve technology’s most vexing problems since 2003, Kove recently launched the world’s first patented and mature software-defined memory (SDM) solution. Kove:SDM™ lets individual servers draw additional memory on demand from a common memory pool. Companies using Kove:SDM™ enjoy faster processing, less energy consumption, reduced costs, and unlimited flexibility. The Kove:SDM™ operator lets companies benefit from Kove’s transformational technology within their Red Hat OpenShift clusters. Potential Kove:SDM™ on Red Hat OpenShift use cases span a wide variety of increasingly memory-demanding applications, including artificial intelligence (AI), machine learning (ML), high-performance computing (HPC), ERP systems, and much more. Kove:SDM™ with Red Hat OpenShift AI use cases span almost every industry.

Benefits:

• Reduced energy costs by as much as 54% while applications ran faster
• Provided unlimited memory, delivering increased flexibility, enhanced performance, and financial savings
• Optimized value of hardware investments, halving hardware capex costs while achieving up to a 100-fold increase in container density

Beth Rothwell, Director of GTM Strategy, Kove: Modern workloads often require more from memory than their hardware can deliver. AI/ML, HPC, ERP systems, animations, high-resolution imaging, and other critical applications require vast amounts of memory for their ever-larger dataset processing and increasingly memory-intensive applications.

Dan Doyle, Business Operations, Kove: Today’s companies are often limited by the amount of memory built into the underlying hardware. But we’re seeing a rise in AI and ML, along with an increase in edge computing, containerization, and other memory-intensive technologies. That built-in memory is struggling to scale at the same rate. A containerized AI application, for instance, would go to storage or fail if its worker node didn’t have enough memory available. And that would either slow or kill the application.

Doyle: Kove:SDM™ is a truly revolutionary memory virtualization software solution that lets individual servers draw from a common memory pool to receive the exact amount of memory they need in real time. I’ve actually heard computer science professors say that memory virtualization is impossible because of the laws of physics. There’s no need to debate the laws of physics here, but the science of memory virtualization is possible. We’ve cracked it with Kove:SDM™.

Kove:SDM™ is made up of 3 software components: a management console (MC) that orchestrates memory pool usage and where users can set up their own provisioning rules; the software that connects applications to the memory pool; and the software that converts servers to the memory targets that, together, form the memory pool.

When an application needs additional memory, it reaches out to the memory pool. The memory pool allocates the additional memory it needs. It then continues to run without interruption. And it doesn’t matter if the amount of memory it needs is larger than the memory within the physical server it’s running on. We’ve seen Kove:SDM™ deliver memory that’s 1,200 x the memory available in the physical server an application was running on.

Rothwell: Our software works anywhere Red Hat Enterprise Linux® or Red Hat OpenShift works, because we designed Kove:SDM™ around Red Hat technologies. It’s particularly useful for clients running Red Hat OpenShift and AI use cases, whether AI model design, training, or inferencing—model training in particular—because it requires extensive amounts of data processing, and that data processing is evolving.

The Kove:SDM™ operator is available on the Red Hat Ecosystem Catalog. We’ve automated the installation process, so companies can download and integrate it with a Red Hat OpenShift cluster in just 4 simple steps.

Doyle: It doesn’t matter whether companies are running Red Hat OpenShift on bare metal, Red Hat Enterprise Linux®, or Red Hat OpenStack® Platform. When Red Hat OpenShift runs with Kove:SDM™, any container workload—AI or otherwise—is not limited by the memory available on the node. The memory pool sits on a dedicated Red Hat Enterprise Linux server. We chose Red Hat Enterprise Linux because it’s widely adopted, vendor-neutral, and very stable, and Red Hat provides regular updates and enterprise-level support.

Rothwell: The Kove operator, in combination with Red Hat OpenShift, provides unlimited memory, which provides increased flexibility, enhanced performance, and financial savings. Red Hat and Supermicro stress-tested Kove:SDM™ and Red Hat OpenShift, testing 2,165 quadrillion results without changing any code. Applications ran faster, while energy costs came down by as much as 54%.

When you consider the enormous amount of energy required to train AI models, 54% is dramatic. Imagine cutting a datacenter power draw in half. As these models grow and develop, the memory and energy they require will grow, too. Using Kove:SDM™ and Red Hat OpenShift to overcome server memory limitations will make a big difference in terms of cost. It’s an important consideration for all organizations that have ESG requirements, especially if they need to run memory-intensive applications.

Doyle: The combination of Kove:SDM™ and Red Hat OpenShift allows organizations to optimize the value of their hardware investments. The testing done by Red Hat and Supermicro achieved up to a 100-fold increase in container density. Customers don’t need additional servers to overcome memory limitations, which the tests showed to halve hardware capex costs.

Kove:SDM™ and Red Hat OpenShift in combination not only brings about financial savings but also dramatically increases flexibility. Developers—and AI model developers, in particular—no longer need to spend time segmenting data to fit it inside servers’ memory, freeing them to spend more time working on their models.

Rothwell: A wide range of use cases would benefit from Kove:SDM™. I mentioned some of the types of increasingly memory-intensive applications we’re seeing today in addition to AI/ML use cases, including HPC and ERP systems.

The list of industry use cases is endless. Biotechnology and bioinformatics spring to mind, specifically for drug development and disease diagnosis. Using Kove:SDM™ with OpenShift AI means they can fully utilize their immense datasets while staying on premise. 

Another example is financial services. One of our customers required a scalable but cost-effective approach to memory to pursue new AI modeling and pre-validation for money transfer and fraud detection worldwide. Kove:SDM™ and Red Hat OpenShift together delivered as much memory as needed, well beyond server limits.

Doyle: There are also several use cases from an operations standpoint. Take disaster recovery: Kove:SDM™ can step in automatically to provide memory in the event of a Dual In-line Memory Module (DIMM) failure, avoiding downtime. In addition to providing an alternative to buying additional servers when on-premise servers reach their memory limitations, it can also provide an alternative to costly cloud bursting—especially for companies needing to stay on premise for compliance reasons.

Rothwell: We’ve been partners since 2015, and that partnership has really accelerated these past 6 months at both a business and technical level. Take the stress testing mentioned earlier: Red Hat led that initiative, and it has provided Kove with some invaluable metrics to demonstrate the value of our technology.

We recently had Red Hat experts join our CEO, John Overton, on stage for keynote speeches at 3 summits, demonstrating the strength of the partnership. We’re also presenting internally to Red Hat teams to help them understand our technology and the value it brings so they can share its value with any of their customers who are being held back by memory limitations.

Doyle: Working with the technical team at Red Hat means we can ensure our technology is available for the most recent versions of Red Hat technologies—and, indeed, new, emerging Red Hat technologies. Being able to connect with Red Hat’s expertise is really important for us.

Doyle: Our technology is evolving with the Red Hat portfolio, so we’re focusing on delivering our SDM for Red Hat OpenShift Virtualization and Open Data Hub at the moment. The servers running the applications using Kove:SDM™ can run not only on bare metal but also containers or, in the near future, using OpenShift Virtualization. We’re also exploring Kove:SDM™ for Red Hat OpenShift running in the cloud.

Rothwell: We’ve recently launched a Kove:SDM™ Memory Tower in partnership with Viking Enterprise Solutions (a Sanmina company), Red Hat, and Computacenter. The tower provides a ready-built memory pool, giving customers plug-and-play access to our virtualized memory solution.